There have been more than 500,000 opioid overdose deaths since 2000. To analyze the opioid epidemic, we construct a model where individuals, with and without pain, choose whether to misuse opioids knowing the probabilities of addiction and dying. These odds are functions of opioid use. Markov chains are estimated from the US data for the college and non-college educated that summarize the transitions into and out of opioid addiction as well as to a deadly overdose. We construct a structural model that matches the estimated Markov chains. We also examine the epidemic's drivers and the impact of medical interventions.